fixtures.reorder_items performs poorly for large numbers of parametrized fixtures

[cheezman34]

If a parametrized fixture generates n items, then reorder_items will have roughly O(n^2) performance. Slice items is called for each item and then iterates over the list of items to find matching groups. We can improve this by pre-caching items by fixture (like argkeys_cache but in reverse).

I've already got code ready for a pull request, I'm just making this issue for record-keeping / discussion.

[pytestbot]

GitMate.io thinks the contributor most likely able to help you is @RonnyPfannschmidt.

[RonnyPfannschmidt]

thanks for doing all the things in a very detail oriented manner, for personal reasons i wont be able to give it the time it deserves until next week

[edmorley]

For anyone who arrived here from the changelog, the fix for this landed in #3108.
(This issue can now be closed)

[nicoddemus]

Thanks @edmorley!